minke.yu/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 12 Packages Projects Releases Wiki Activity

[Fix] Fix gpt oss triton kernels and upgrade flashinfer back to 0.6.11.post1 (#25335)

Browse Source 
Co-authored-by: sglang-bot <sglang-bot@users.noreply.github.com>
Co-authored-by: b8zhong <b8zhong@users.noreply.github.com>
Co-authored-by: mmangkad <mmangkad@users.noreply.github.com>
This commit is contained in:
Baizhou Zhang
2026-05-15 01:04:56 -07:00
committed by GitHub
co-authored by sglang-bot b8zhong mmangkad
parent 4adfc6cf7e
commit 0c19540550
13 changed files with 147 additions and 53 deletions
Download Patch File Download Diff File Expand all files Collapse all files
test/registered/moe/test_cutedsl_moe.py
+9 -6
View File
@@ -899,13 +899,14 @@ class TestCuteDslV1(unittest.TestCase):
masked_m.to(hidden_states.device),
)
a_global_scale = input_global_scale[:1]
a_fp4, a_scale_interleaved = fp4_quantize(
hidden_states, input_global_scale
hidden_states, a_global_scale
)
a_in_dtype = dequantize_nvfp4_to_dtype(
a_fp4,
a_scale_interleaved,
input_global_scale,
a_global_scale,
dtype=hidden_states.dtype,
device=hidden_states.device,
block_size=16,
@@ -1077,11 +1078,12 @@ class TestCuteDslV1(unittest.TestCase):
masked_m.to(device),
)
a_fp4, a_scale_interleaved = fp4_quantize(hidden_states, input_global_scale)
a_global_scale = input_global_scale[:1]
a_fp4, a_scale_interleaved = fp4_quantize(hidden_states, a_global_scale)
a_in_dtype = dequantize_nvfp4_to_dtype(
a_fp4,
a_scale_interleaved,
input_global_scale,
a_global_scale,
dtype=hidden_states.dtype,
device=device,
block_size=16,
@@ -1251,11 +1253,12 @@ class TestCuteDslV1(unittest.TestCase):
)
# PyTorch reference (same as the bf16 input test)
a_fp4, a_scale = fp4_quantize(hidden_states, input_gs)
a_gs = input_gs[:1]
a_fp4, a_scale = fp4_quantize(hidden_states, a_gs)
a_deq = dequantize_nvfp4_to_dtype(
a_fp4,
a_scale,
input_gs,
a_gs,
dtype=torch.bfloat16,
device=device,
block_size=16,